I wasn't sure what conditions could cause a "partial write." If it's only a disk-full condition, and not a hardware-buffer-full condition or something, then syswrite will probably do what I want in all the cases I care about. Thanks!

The alternative method you suggest is the solution I'm currently using. It works just fine, but I don't like having to reopen the files all the time. I'm probably just picky :)

Thanks,
Alan

Update: Ah, sorry for the misunderstanding. Someone else suggested the same alternative in another thread (open each file only once, and flock many times, in each child).

I think the best solution is a combination of that, with a redistribution of the problem space across my child processes. Instead of sending each child only one sub-problem and then letting it die, I could avoid one fork per sub-problem by giving each of my N children 1/Nth of the problem space all at once. If I do that, opening the file once per child will make a lot more sense, because each child will be doing a lot more writing. As it is now, each child only writes a few lines, and does not always write to every possible file, so there may be more opens if I open every file once per child, than if I continued to open each file only when I append a line :)

Thanks!